Reply #107: Clarification on random sampling

[Febble]

for the sake of any unquiet spirits who may be hovering....

Voters may (or may not) have been selected at random. The protocol should have ensured random selection if it was adhered to and if fluctuations in coverage did not also coincide with clusters of like voters.

There is some evidence that selection was not random, but yes, selection was certainly intended to be random.

However, response rate is not random. It depends on voter characteristics. Voters do not toss a coin when deciding whether or not to participate in a poll. They simply decide.

The reason that age, sex and race of both respondents and non-respondents are recorded, is so that if the sample of respondents contains a different proportion of voters with these characteristics than the sample of selected non-responders, the respondent sample can be weighted to match. I have no problem with that; I believe TIA has no problem with it either.

And because this reweighting is necessary, we know that non-response bias exists. For example, it appears that men are more likely to refuse than women.

The problem is that the pollsters can only weight for visible characteristics. They know if they have a reluctant male responder problem; or a reluctant white responder problem; or a reluctant middle-aged responder problem; they do not know if they have a reluctant Bush responder problem; even less do they know whether they have a reluctant Gore defector problem.

But to assume that the sample was random flies in the face of the evidence that weighting was required for visible characteristics. Even if selection was random, response wasn't, and never is.

Which is where the binomial theorem bites the dust.